stanford-mast
Stanford MAST conducts research in computer architecture and computer systems to make computing systems at all scales faster, more energy-efficient, cost-effective, and secure. Its BLAST project provides open-source VM and sandbox orchestration through a small REST API and a single binary.
At a Glance
- AI-agent and browser-automation developers
- Coding-agent platform builders
- Developers and enterprises running local or bring-your-own-cloud sandbox compute
- Research and benchmarking teams evaluating browser agents
- +1 more
AI Tools by stanford-mast
(1)
BLAST
VM Sandbox Orchestration Binary
Discussions
No discussions yet
Be the first to start a discussion about stanford-mast
Latest News
BLAST highlighted as a 7 MB MIT-licensed local sandbox-orchestration binary for VMs, with REST APIs and SmolVM/Hypeman/Docker abstraction.
Latest recorded BLAST repository Publish commit by Caleb Winston.
BLAST README updated during ongoing project development.
Browser configuration updated to use browser profiles, with updated ChatOpenAI and BLAST usage guidance.
Products & Services
An open-source, MIT-licensed VMs-as-a-service and local/BYOC sandbox-orchestration system. The current distribution is a single approximately 7 MB binary, installable with Cargo, that exposes a REST API for forking VMs, running commands, syncing files, managing sessions, and monitoring VM/run state; it can orchestrate SmolVM, Hypeman, and Docker-backed sandboxes.
Market Position
BLAST positions itself against local sandboxing and BYOC sandbox-compute solutions that provide isolated execution utilities but not full orchestration. Its differentiators are a small single binary, permissive MIT license, provider abstraction across SmolVM/Hypeman/Docker, and built-in scheduling, placement, forking, snapshots, synchronization, migration, and resource management; this favors self-hosted control and avoids dependence on a hosted, metered sandbox service.
Founding Story
BLAST was started as an open-source effort by Stanford MAST researchers to make browser-augmented AI and sandbox compute practical at scale. The initial 2025 vision was a high-performance serving engine for web-browsing LLMs that could approach search-engine-level latency through automatic parallelism, prefix caching, budgeting, and an OpenAI-compatible API; the project subsequently evolved toward general local/BYOC VM orchestration.
Business Model
Revenue Model
BLAST is distributed as free, open-source MIT-licensed software; the public materials describe self-hosted/local and BYOC deployment rather than subscriptions, usage billing, or paid pricing tiers.
Target Markets
- AI-agent and browser-automation developers
- Coding-agent platform builders
- Developers and enterprises running local or bring-your-own-cloud sandbox compute
- Research and benchmarking teams evaluating browser agents
- Cloud-infrastructure and systems researchers
- Background coding-agent platforms requiring local VMs or VMs on BYOC machines
- Sandboxed command execution and isolated development workloads
- Forking VM state for feature branches or parallel tasks
- Browser-augmented LLM and web-browsing AI applications
- Deployments needing a unified control plane across bursty cloud compute and user-provided compute
- Custom browser-agent task evaluation and benchmarking